A RAG chatbot that knows your docs by heart.
RAG-based on your PDFs, Excel sheets and full websites. Live handover to your team, WhatsApp Business + HubSpot built in, fluent German and English with formal/casual register switch. Over 156 features — built for enterprise depth, not chatbot toys.
Four breaks that make generic bots useless in practice.
The market is full of chatbot tools that paste a generic LLM onto your website. Pretty in the demo, useless in production. Here are the four gaps a serious bot in 2026 has to close.
Answers from the void.
Standard bots pull answers from training data — i.e. from whatever was on the public web months ago. Ask about your product datasheet? Hallucination. Ask about your office hours? Hallucination. When the bot knows nothing about YOU, it hurts more than it helps.
"Hello, how can I help you?" and not much more.
English-trained models treat German (and most non-English languages) as a translation layer. Answers come out stiff, awkwardly polite, often technically wrong. They can't hold a register (formal vs. casual) consistently and mistranslate industry terms. For DACH buyers that's a direct trust signal: this vendor doesn't take German seriously.
The bot doesn't know — and nobody hears about it.
When the bot can't answer it says "Sorry, I can't help with that" and the user leaves. No lead capture, no escalation to your team, no CRM record. One lead per day quietly lost because the bot can't route.
Every inquiry gets treated the same.
A pricing question, a complaint, a booking request, a sales lead — all land in the same funnel. Generic bots don't classify. Structured routing logic (pricing → sales pipeline, complaint → support queue, lead → CRM with score) is missing in most tools.
What you actually get.
Webweezl AI Chatbot is three disciplines in one system: RAG knowledge base from your own documents, fluent multilingual output with proper register handling, and a live handover workflow with lead capture. Over 156 features in 17 categories — that's the honest count, not marketing math.
The bot learns from your sources — not from the open internet. You feed it, it answers from that knowledge.
Upload as many documents as you want — datasheets, manuals, FAQs, contracts. Automatic chunking, vector indexing, updates without a full re-index.
Drop in your domain, the bot recursively crawls and uses your full site as the knowledge base. When content changes, only the changed pages are re-indexed. Sitemap-aware.
When RAG isn't enough: define guided conversations visually — per industry, per use case. Conditional branches, variables, skip logic. You build the flows, the bot follows them.
Optimized for German and English in parallel, with proper register handling. Sounds like an employee answering — not a translated US tool.
Per-bot setting: formal address (German Sie / English you-with-distance) or casual (Du / first-name basis). Consistent across the whole conversation, including long sessions. No accidental tone-switch mid-stream.
Inquiries get classified: product question, pricing, booking, complaint, lead inquiry, etc. 8-class routing engine, 17 sub-categories for finer logic. That's the basis for structured CRM handling.
German and English in parallel by default. The bot auto-switches to the user's language. Answer content without translation loss — both languages generated directly from the source knowledge.
When the bot can't answer — or when the lead is ripe for a real conversation — it hands over cleanly. With full conversation history, with CRM record creation, with Slack/Teams notification.
Mid-conversation the bot can hand over to a human teammate. The agent sees the full history, takes over seamlessly. Spectating mode (agent reads along without taking over). Canned responses. Reassign logic.
On a qualified inquiry the bot captures name, email, company — structured, not pushy. Data flows directly into HubSpot (native) or via webhook into any other CRM. Conversation excerpt attached as a CRM note for sales.
Same bot reachable over the WhatsApp Business API. Conversation can start on the site, continue on WhatsApp, all in the same knowledge context. Central inbox dashboard for your team.
Not "another chatbot". An employee that fits your knowledge.
Most chatbots fail at one of two points: they don't know your products (no RAG, just training data) or they can't hand over to a human (dead end instead of escalation). Webweezl AI Chatbot is built differently at exactly those two points.
RAG with your documents. Vector database in Cloudflare Vectorize, inference via Gemini 2.5 Flash Lite. Each answer is a synthesis of 3–7 knowledge chunks pulled from your own documents — with source attribution. When a user asks "which ink colours do you stock?", the bot quotes page 12 of your product catalog, not generic internet knowledge.
Human Takeover is deliberately gated. Free and Starter users see that the bot detects an escalation moment — but can't respond as a human yet. That's an intentional upgrade path: a clear "I need this" trigger for the Business tier. On Business+ the full handover runs with notifications, spectating, canned responses, reassign logic. This isn't an add-on — it's built like Intercom or Drift, but as an integrated layer of a RAG bot.
That's why 156+ features in 17 categories — not because we needed marketing-features, but because a real enterprise bot genuinely needs all of it.
- 01 Cloudflare Workers + Vectorize Edge compute with EU region. Vector DB for RAG retrieval. Sub-200ms response time across DACH.
- 02 Gemini 2.5 Flash Lite Primary LLM backend, optimized for speed + context depth. Migration to Gemini 3.1 Flash Lite (with Thought Signature Circulation) planned for Q2 2026.
- 03 156+ features in 17 categories Visual flow editor (Drawflow-based), answer validation, session memory, entity tracking, conversational forms, lead management, analytics dashboard, website crawler, multi-tenant architecture.
- 04 Human Takeover · Business+ Agent notifications, typing indicators, spectating, reassign logic, canned responses. Built like standalone tools (Intercom-grade), integrated into the RAG flow.
- 05 5-tier RBAC + 2FA Granular team-role structures, TOTP-based two-factor authentication. Enterprise-grade security.
- 06 WhatsApp Business + HubSpot Native Official WhatsApp Business API integration, native HubSpot sync (not via Zapier). Plus generic webhook interface for other CRMs.
Who uses Webweezl AI Chatbot to its full potential.
Built for DACH SMBs through enterprise — companies with high volumes of repetitive inquiries who already have the knowledge documented somewhere, just not findable enough.
DACH SMBs with complex product portfolios
Machinery, tools, specialty chemicals, industrial hardware. You have a catalog of 200+ products, each with their own specs. Customers ask “does model X work with model Y?”. Manual answering is impossible, a generic bot lies. A RAG bot pulls from your catalog and answers exactly.
▶ Inquiry volume halved without adding headcountEducation / higher-education institutions
Programme advice, course inquiries, application processes. Applicants ask constantly “do you accept my qualification?”, “which programme fits my background?”, “how does the application work?”. The bot knows your programme descriptions, can escalate, can route to advising slots.
▶ Programme advisory inquiries captured structurallyB2B SaaS with deep documentation
SaaS vendors with help centres, setup guides, API docs. Instead of every setup question becoming a support ticket: the bot pulls from the docs, gives a structured answer, escalates only when there's real need. Reduces support load, accelerates time-to-value.
▶ Support ticket volume reduced · onboarding acceleratedE-commerce with complex pre-sale advice
Furniture, tools, premium D2C. Customers need advice BEFORE buying — does the sofa fit my living room? Which size? The bot knows your inventory, asks back, recommends structurally. On complex inquiries: handover to a human.
▶ Conversion rate on premium products liftedSee pricing on the product site.
Webweezl AI Chatbot has its own marketing site with the up-to-date pricing table, per-tier feature comparison, free-trial sign-up and annual toggle. Rather than duplicating the table here (where it'd go stale) — go straight to the source:
See pricing on webweezl.io
4 tiers — Free, Starter, Business, Enterprise. Full feature comparison per tier, annual discount, sign-up directly on the site. 14-day free trial available.
▸ webweezl.io/#pricing ↗ Prefer a conversation? Book a direct slot →What you should know upfront.
The questions that come up again and again in first-call conversations about bot setups.
-
01 How is this different from a ChatGPT plugin or Intercom? +
Three differences: (1) RAG with your documents. ChatGPT plugins have no grounding in your data — they guess. Webweezl AI Chatbot pulls from your PDFs, Excel and site with source attribution. (2) First-class multilingual. Optimized for German + English with proper register handling, not just English-to-German pass-through. (3) Human Takeover is built in, not a separate tool. Intercom is good at live chat, weak at RAG. ChatGPT is the inverse. Webweezl bundles both natively.
-
02 Can the bot learn from my own documents? +
Yes — that's the whole point. PDFs, Word, Excel, images, full websites — everything gets indexed into a vector database (Cloudflare Vectorize). The bot answers from exactly that knowledge, with source attribution. When documents change, only the changed sections are re-indexed.
-
03 Does the bot work on my WordPress / Shopify / Webflow / custom site? +
Yes — integrates as a JavaScript widget (one snippet in the footer) or as an iframe embed. Works in WordPress, Shopify, Webflow, Astro, Next.js, custom themes. Setup is typically 30 minutes from snippet drop-in to first answer.
-
04 How long until the bot knows my site and documents? +
Initial indexing: ~30 minutes for 50–100 documents, ~1–2 hours for full sites with 500+ pages. Updates are incremental — new documents are available within minutes.
-
05 Can the bot book appointments or send leads to my CRM? +
Yes. Native HubSpot integration plus webhook-out for any other CRM (Pipedrive, Salesforce, custom APIs). On a qualified inquiry the bot captures lead data structurally, creates the CRM record, and attaches the conversation history as a note. Calendar booking via Webweezl Connect or external calendars (Google Workspace, Outlook).
-
06 What about GDPR and data residency? +
Cloudflare Workers run with EU region routing. Vector DB (Cloudflare Vectorize) in European edge data centres. Inference via the Gemini API (Google EU region). No data in the US, full data-processing-agreement (DPA) documentation. Self-hosted fonts, no tracking pixels without consent.
-
07 How fast are the answers? +
Standard response time: 2–4 seconds across DACH (Cloudflare Edge EU + Gemini EU region). Streaming is on by default: the user sees the first words after about 800ms, the full answer within 4 seconds.
-
08 What happens when the bot doesn't know? +
Three paths, configurable per setup: (a) Honest "I can't answer that — would you like to talk to someone?" with lead capture. (b) Live handover to your team (Slack/Teams notification with conversation history). (c) Routing to FAQ section or contact page. What doesn't happen at Webweezl: the bot hallucinates an answer to avoid saying "I don't know".
-
09 Do I get data on bot performance? +
Yes — admin dashboard with conversation volume, top inquiries, escalation rate, lead-conversion, source-attribution stats. Weekly performance email (optional). All conversations exportable as CSV.
Five platforms. One hand.
Webweezl AI Chatbot is one of five SaaS products I've built and run myself.
If you want to use the bot as part of a larger marketing setup — with lead routing, CRM integration, performance tracking — we can look together at Marketing Automation and CRM & Systems.